{"id":"d5d6af49-f258-478b-987c-d69b46b4206b","arxiv_id":"2506.02340","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An explicit integral formula for the heat kernel and the full Laplace spectrum of the Cayley graph of the modular group are derived by solving the spectral problem on a weighted line that the graph covers.","lead":"The authors write down an explicit formula for the heat kernel on the Cayley graph of the modular group PSL2(Z), the infinite graph made from a generator of order 2 and one of order 3. The formula yields the full Laplace spectrum, and the paper links its lowest value to a conjecture about the spectral gaps of finite quotients PSL2(Fp), in the spirit of Selberg's 1/4-conjecture.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The negative-n branch of Theorem 1.1 is off by a factor: the lifting step uses a fiber-size identity that is false for n<0, and formula (1.5) contradicts the defining heat equation at x=ab.","rationale":"The reader identified the same weak point: the lifting step for negative n is not justified. The new stress-test analysis shows the issue is not merely a missing domain clause; it produces a concrete numerical contradiction with the heat equation. For x = ab, pi(x) = -2, and the heat kernel must satisfy (4.3) with the Laplacian (1.2). A two-line random-walk computation gives k_t(ab) = t^2/16 + O(t^3). The paper's own projected-line data give h^pr_t(0,-2) = t^2/(8 sqrt(2)) + O(t^3), and the paper's prefactor sqrt(2)^(-ceil(-1)) = sqrt(2) multiplies this to t^2/8, a factor of two too large. The correct prefactor is 1/sqrt(2), reflecting the true fiber size 2 = |pi^{-1}(-2)|. Therefore Theorem 1.1, the central heat-kernel claim, is false as printed. The spectrum in Corollary 1.1 is a support statement and may well be correct, and the n >= 0 branch may also be correct, but the main theorem's negative-n formula fails. Since the abstract and introduction advertise the explicit formula without restriction, the current version should not be accepted without correction. The verdict is moved to REJECT: the manuscript needs a substantive revision of the negative-n normalization and of the ancillary fiber-size statement before the central claim can be evaluated.","tokens_in":21505,"tokens_out":28793,"duration_ms":288442,"concrete_test":"Compute the t^2 coefficient of k_t(ab) directly from the series e^{-tL}delta_e with L in (1.2); it is 1/16. Then evaluate the right-hand side of (1.5) at n = -2, either by expanding the proof's expression K_t(-2) = sqrt(2)^(-ceil(-1)) h^pr_t(0,-2) with h^pr_t(0,-2) = (1/(8 sqrt(2)))t^2 + O(t^3), or by numerical integration of (1.5). If the coefficient is 1/8 rather than 1/16, the negative-n prefactor is incorrect; replacing it by 2^(-floor(|n|/2)/2) should restore agreement.","verdict_should_be":"REJECT","load_bearing_attack":"The proof of Theorem 1.1 sets K_t(n) = sqrt(2)^(-ceil(n/2)) h^pr_t(0,n), citing |pi^{-1}(n)| = 2^(ceil(n/2)) (Section 5.4). This identity is valid only for n >= 0. For n < 0, a reduced word beginning with a has length N = -n and exactly 2^(floor(N/2)) choices: the letters in even positions are b or b^{-1}, and all other positions are forced. For example, pi^{-1}(-2) = {ab, ab^2} has size 2, while 2^(ceil(-2/2)) = 1/2. Hence the prefactor for n < 0 should be 2^(-floor(|n|/2)/2), not sqrt(2)^(-ceil(n/2)). The error is quantitative. For n = -2, the defining equation (4.3) with L from (1.2) gives k_t(ab) = (1/16)t^2 + O(t^3): the first-order term vanishes, and the two-step path ab -> a -> e has probability (1/4)(1/2) = 1/8, so the t^2 coefficient is P^2(ab,e)/2 = 1/16. On the projected line, (5.3)-(5.4) give P^pr(0,-1) = 1/2 and P^pr(-1,-2) = 1/(2 sqrt(2)), hence h^pr_t(0,-2) = (1/(8 sqrt(2)))t^2 + O(t^3). The printed formula K_t(-2) = sqrt(2) h^pr_t(0,-2) then yields t^2 coefficient 1/8, twice the correct value; the correct lifting coefficient is 1/sqrt(2). Thus Theorem 1.1 as stated is false for words beginning with a. Corollary 1.1 may survive because it depends only on the support of the spectral measure, but the explicit heat kernel formula is wrong on the negative-n branch.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Cayley graph Γ of G = PSL₂ℤ ≅ C₂ ∗ C₃ with generators a (order 2, realized as a double edge) and b (order 3), together with the normalized Laplacian L defined in (1.2). The main result, Theorem 1.1, claims an explicit formula k_t(x) = K_t(π(x)) for the heat kernel, where π(x) is the signed word length from (1.4). The proof proceeds by viewing Γ as a strong regular covering of a weighted infinite line ℤ, solving the spectral problem for the projected Laplacian L^pr on ℤ (Theorem 5.1), and then lifting the line heat kernel back to Γ via the Chung–Yau transfer formula. Corollary 1.1 identifies the spectrum of L as two intervals together with the isolated points 3/4 and 7/4, with bottom λ₀ ≈ 0.01234, and the paper conjectures that λ₀ bounds the spectral gap of Cayley graphs of PSL₂𝔽ₚ with the same generators, with numerical evidence for small primes.","tokens_in":21794,"tokens_out":26687,"duration_ms":262948,"significance":"If Theorem 1.1 is corrected as indicated below, this is a substantial contribution: it gives an explicit heat kernel and spectrum for an infinite non-tree Cayley graph with torsion, obtained through a self-contained spectral resolution of a weighted line and a rigorous covering transfer argument. The main strengths are the closed-form spectral theorem for L^pr, the parameter-free derivation of the constants λ₀, λ₁, 3/4 and 7/4, and the explicit machine-checkable coefficients in the heat-kernel formula. The conjecture connecting λ₀ to Selberg's 1/4-conjecture is interesting and is supported by computations, even though it remains a conjecture.","major_comments":[{"comment":"The lifting step uses the identity |π^{-1}(n)| = 2^{⌈n/2⌉}, but this identity is valid only for n ≥ 0. For n < 0 the fiber size is |π^{-1}(n)| = 2^{⌊|n|/2⌋}; for example, π^{-1}(-2) = {ab, ab²} has two elements, while 2^{⌈-2/2⌉} = 1/2. Consequently the formula K_t(n) = √2^{-⌈n/2⌉} h^pr_t(0,n) is wrong on the negative branch. The correct prefactor for n < 0 is √2^{⌈n/2⌉} (the reciprocal of the printed one), and therefore the displayed coefficients α_n, β_n and γ_n^±(s) for n < 0 in (1.5) are also incorrect as printed: they should be multiplied by 2^{⌈n/2⌉}, equivalently the prefactor inside γ_n^± should be √2^{⌈n/2⌉} instead of √2^{-⌈n/2⌉}.","section":"Section 5.4, proof of Theorem 1.1"},{"comment":"The error is quantitative and falsifies Theorem 1.1 as stated. For x = ab, which has π(ab) = -2, the heat equation (4.3) with L from (1.2) gives k_t(ab) = (1/16)t² + O(t³): the only two-step path from e to ab is e → a → ab, with transition probabilities 1/2 and 1/4, contributing (1/2 · 1/4)/2 = 1/16. On the projected line, (5.3)–(5.4) give L^pr(0,-1) = -1/2 and L^pr(-1,-2) = -1/(2√2), so h^pr_t(0,-2) has t²-coefficient 1/(8√2). The printed prefactor √2 multiplies this to 1/8, which is twice the correct value. Thus formula (1.5) contradicts the defining heat equation for words beginning with a. Since Corollary 1.1 uses Theorem 1.1 as the upper-bound input, that proof is not complete as written; however, the error is local and the corrected prefactor still yields a heat kernel with spectral support in Sp(L^pr), so the spectral conclusion is plausibly recoverable.","section":"Theorem 1.1, n < 0 branch; equation (4.3)"}],"minor_comments":[{"comment":"The sewing equation uses ̊check{v}_{-1}, but ̊check{v}_n was defined only for n ≥ 0; the notation should be adjusted, for example by writing f(0) = u₀ directly instead of ̊check{v}_{-1}.","section":"Section 5.3, equation (5.7)"},{"comment":"The sentence 'In the formulas with double lines below, the first line corresponds to even n = 2m, and the second line to odd n = 2m + 1' should specify whether m ranges over all integers or only nonnegative integers when n < 0; this matters for the explicit evaluation of sin(ms) for negative m.","section":"Theorem 1.1, displayed formulas"},{"comment":"The title contains a typo, 'HEA T KERNEL' should be 'HEAT KERNEL'.","section":"Title"}],"recommendation":"major_revision","confidential_remarks":"The negative-branch prefactor error is significant but appears to be a localized, fixable mistake: the correct prefactor for n < 0 is the reciprocal of the printed one, and the spectral support argument should survive the correction. I recommend major revision rather than rejection, with the request that the authors correct the prefactor, update the explicit coefficients for n < 0, and verify the small-time expansion at several negative vertices such as ab and aba."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the explicit heat kernel formula in Theorem 1.1 is wrong for negative n, and the error is not a minor typo. The paper's main new claim fails as stated; the line spectral analysis and the covering framework are solid and worth preserving.\n\nWhat is actually new: the paper solves the spectral problem on the weighted line explicitly, computes the line heat kernel, and then attempts to lift it via a strong regular covering. The line part is well done; I checked the Laplacian coefficients (5.2), the discrete eigenvalues 3/4 and 7/4, and the band edges, and they are internally consistent. Corollary 1.1 matches Cartwright–Soardi, which the paper acknowledges. The PSL2(Fp) conjecture is interesting, though the numerical evidence as given is not reproducible from the text.\n\nThe soft spot is a genuine error. The proof of Theorem 1.1 sets K_t(n) = sqrt(2)^{-ceil(n/2)} hpr(0,n), citing |pi^{-1}(n)| = 2^{ceil(n/2)}. That identity holds only for n >= 0. For n < 0, the fiber over -N consists of reduced words starting with a, and there are 2^{floor(N/2)} of them, not 2^{ceil(n/2)}. For n = -2, the printed prefactor is sqrt(2), while the correct lifting factor is 1/sqrt(2). The heat equation check nails it: for x = ab, the true k_t(ab) has t^2 coefficient 1/16 (path ab -> a -> e with weights 1/4 and 1/2), while the printed formula gives 1/8. So Theorem 1.1 is false for words beginning with a. This is not a missing domain note; it is a factor-of-two (or worse, growing with |n|) error in the central formula.\n\nCorollary 1.1 may survive because the spectrum of the covering is inherited from the base, and the line spectrum is correct; but the proof as written relies on the erroneous heat kernel lifting, so it needs repair. The completeness proof of Proposition 5.4 is also compressed—only one mixed-parity case is shown, and the other cases are asserted.\n\nWho is this for: researchers working on explicit spectra of infinite graphs, especially Cayley graphs of amalgams or free products. The method is worth knowing, but as it stands the paper's main result is not usable. It deserves referee time because the spectral part is serious and the error is localized and fixable; a revised version with corrected prefactors (and reproducible numerics) could be a solid paper.\n\nRecommendation: send it to review, but the referee should require the authors to fix the negative-n lifting formula and verify it against the heat equation before acceptance.","headline":"The explicit heat kernel formula is wrong for negative n; the line spectral analysis is solid, but Theorem 1.1 as stated fails.","tokens_in":22479,"tokens_out":9539,"would_cite":false,"duration_ms":93116,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C25","05C50","20F65"],"pacs":[],"model":"deepseek-v4-flash","headline":"The heat kernel of the PSL(2,Z) Cayley graph is written explicitly as two exponentials plus two integrals, and the Laplacian spectrum is the union of two intervals and two isolated points.","keywords":["heat kernel","Cayley graph","PSL(2,Z)","modular group","normalized Laplacian","spectral gap","weighted graph covering","free product C2*C3"],"falsifier":"Count the reduced words of each length that map to a fixed negative integer under $\\pi$; for $n=-2$ the fiber has 2 elements rather than $2^{\\lceil -2/2\\rceil}=1/2$, so the printed normalization fails unless a separate negative-$n$ identity is supplied. Alternatively, expand $k_t$ for small $t$ on a word starting with $a$ from the graph's finite-neighborhood data and compare with (1.5).","tokens_in":21144,"feed_emoji":"📐","tokens_out":11191,"duration_ms":105894,"temperature":0.7,"pith_summary":"Taking the modular group with presentation $\\langle a,b \\mid a^2=1, b^3=1\\rangle$, the paper claims an explicit formula for the heat kernel of its Cayley graph. The method projects the graph onto a weighted infinite line, solves the spectral problem there, and lifts the heat kernel back through a strongly regular covering. Corollary 1.1 states that the normalized Laplacian on $\\ell^2(G)$ has spectrum equal to two intervals, $[\\lambda_0,\\tfrac{7}{4}-\\lambda_1]$ and $[\\lambda_1,\\tfrac{7}{4}-\\lambda_0]$, together with isolated eigenvalues $\\tfrac{3}{4}$ and $\\tfrac{7}{4}$, where $\\lambda_0\\approx 0.01234$ is a positive spectral gap. The paper also conjectures that this same $\\lambda_0$ is the limiting lower bound for spectral gaps of the finite Cayley graphs of $\\operatorname{PSL}_2\\mathbb{F}_p$ with the same generators.","feed_headline":"Full Laplacian spectrum found for PSL(2,Z) Cayley graph","feed_subtitle":"Heat kernel becomes two exponentials plus two integrals; spectral gap is about 0.0123.","key_machinery":"The carrier of the argument is a strong and regular covering of weighted graphs: the Cayley graph $\\Gamma$ maps onto a weighted infinite line $L_\\infty$ with vertex set $\\mathbb{Z}$ and weights $w(2n+1,2n+1)=2^{n+1}$ and $w(2n-1,2n)=w(2n,2n+1)=2^{n+1}$. For strong and regular coverings the heat kernel lifts by a fiber-normalized formula, and here it becomes $k_t(x)=\\sqrt{2}^{-\\lceil n/2\\rceil}h^{\\mathrm{pr}}_t(0,n)$ with $n=\\pi(x)$. On the line, the projected Laplacian is solved completely: its continuous spectral part is parametrized by $x\\in[0,\\pi]$ with eigenvalues $\\tfrac{7}{8}\\pm R_x/2$, and its two discrete eigenvalues $\\tfrac{3}{4}$ and $\\tfrac{7}{4}$ have explicit eigenfunctions.","core_discovery":"On the paper's own terms, the discovery is a complete spectral decomposition for the Cayley graph of $C_2*C_3\\cong \\operatorname{PSL}_2\\mathbb{Z}$. The heat kernel is constant on the fibers of the projection $\\pi:G\\to\\mathbb{Z}$ defined by word length and first letter, so $k_t(x)=K_t(\\pi(x))$, and $K_t(n)$ is given by two exponential terms coming from the eigenvalues $\\tfrac{3}{4}$ and $\\tfrac{7}{4}$ plus two integrals over $s\\in[0,\\pi]$ involving $R_s=\\sqrt{25/16+\\sqrt{2}\\cos s}$. The same decomposition identifies the spectrum of the normalized Laplacian on $\\ell^2(G)$ as $[\\lambda_0,\\tfrac{7}{4}-\\lambda_1]\\cup\\{\\tfrac{3}{4}\\}\\cup[\\lambda_1,\\tfrac{7}{4}-\\lambda_0]\\cup\\{\\tfrac{7}{4}\\}$, with $\\lambda_0=0.01234\\ldots$ and $\\lambda_1=1.0675\\ldots$, so the bottom of the spectrum is strictly positive.","pith_inferences":["The same quotient-by-covering strategy could produce explicit heat kernels for other virtually free groups whose Cayley graphs project onto weighted lines with periodic weights, such as free products $C_m*C_n$ with asymmetric generators.","If the finite-field conjecture holds, the fixed-generator Cayley graphs of $\\operatorname{PSL}_2\\mathbb{F}_p$ would have spectral gaps bounded below by $\\lambda_0$ for all but finitely many $p$, giving a concrete expander statement tied to the infinite graph.","The explicit spectral measure may allow quantitative return-probability estimates for the simple random walk on the modular group, a testable numerical prediction not developed in the paper."],"forward_implications":["The continuous-time random walk on $\\Gamma$ has transition probabilities directly computable from a one-dimensional integral plus two exponential terms.","The Laplacian has a spectral gap $\\lambda_0\\approx 0.01234>0$, which gives exponential decay in $t$ of the heat operator away from the identity at the scale set by $\\lambda_0$.","The isolated eigenvalues $\\tfrac{3}{4}$ and $\\tfrac{7}{4}$ come with explicit $\\ell^2$ eigenfunctions obtained by lifting eigenfunctions of the projected line.","The spectrum can be compared with the earlier harmonic-analysis spectrum for free products of cyclic groups, matching the shape of two bands plus two atoms."],"supporting_citations":[{"why":"Supplies the strong-and-regular covering setup and the heat-kernel transfer formula that the paper extends to infinite fibers.","marker":"[5]"},{"why":"Gives the earlier harmonic-analysis spectrum for the free product of two cyclic groups, the comparison target for Corollary 1.1.","marker":"[3]"},{"why":"Provides earlier explicit heat-kernel formulas for regular graphs that motivate the approach.","marker":"[4]"},{"why":"Provides the general method for constructing explicit heat kernels on infinite graphs that this paper applies.","marker":"[9]"},{"why":"Defines the Cayley graph of the modular group as a quotient of the 4-regular tree, fixing the base graph of the covering.","marker":"[17]"}],"fun_headline_variants":["Explicit heat kernel solves PSL(2,Z) Cayley graph","Spectral gap of modular group graph pinned down","Heat kernel formula reveals full spectrum of PSL(2,Z)","Two eigenvalues plus integrals: modular group Laplacian exposed","Analogy with Selberg fuels new spectral gap conjecture"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The lifting step assumes the fiber-size identity $|\\pi^{-1}(n)|=2^{\\lceil n/2\\rceil}$ holds for every integer $n$; the paper prints it without a domain, and for negative $n$ the counts are different, so the normalization of the formula for words beginning with $a$ is not established as written.","fun_headline_variants_meta":{"raw":{"variants":["Explicit heat kernel solves PSL(2,Z) Cayley graph","Spectral gap of modular group graph pinned down","Heat kernel formula reveals full spectrum of PSL(2,Z)","Two eigenvalues plus integrals: modular group Laplacian exposed","Analogy with Selberg fuels new spectral gap conjecture"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000526,"raw_usage":{"total_tokens":2548,"prompt_tokens":962,"completion_tokens":1586,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":578,"completion_tokens_details":{"reasoning_tokens":1514}},"tokens_in":578,"tokens_out":1586,"duration_ms":13578,"temperature":1.0,"reasoning_tokens":1514,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:31:26.635425+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Count the reduced words of each length that map to a fixed negative integer under $\\pi$; for $n=-2$ the fiber has 2 elements rather than $2^{\\lceil -2/2\\rceil}=1/2$, so the printed normalization fails unless a separate negative-$n$ identity is supplied. Alternatively, expand $k_t$ for small $t$ on a word starting with $a$ from the graph's finite-neighborhood data and compare with (1.5).","supporting_citations":[{"cited_title":"and Yau, S.-T., 1999, Coverings, heat kernels and spanning trees, Electron","cited_arxiv_id":null,"evidence_quote":"Supplies the strong-and-regular covering setup and the heat-kernel transfer formula that the paper extends to infinite fibers."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the earlier harmonic-analysis spectrum for the free product of two cyclic groups, the comparison target for Corollary 1.1."},{"cited_title":"and Karlsson, A., 2015, Heat kernels on regular graphs and generalized Ihara zeta function formulas, Monatsh","cited_arxiv_id":null,"evidence_quote":"Provides earlier explicit heat-kernel formulas for regular graphs that motivate the approach."},{"cited_title":"and Smajlovi´ c, L., 2026, Constructing heat kernels on infinite graphs, Anal","cited_arxiv_id":null,"evidence_quote":"Provides the general method for constructing explicit heat kernels on infinite graphs that this paper applies."},{"cited_title":"Math., Springer-Verlag, Berlin","cited_arxiv_id":null,"evidence_quote":"Defines the Cayley graph of the modular group as a quotient of the 4-regular tree, fixing the base graph of the covering."}],"review_version":1}